EntryUpdated 2026/10/25

Money Can Build Fabs. Why Not a Chip Supply Chain?

Because every layer has its own bottleneck, and the bottlenecks sit in different countries. Diversifying design does not offset concentration in manufacturing either: the two are risks at different layers.

The bottlenecks sit in different countries, so no one country can clear them alone

Chip design and EDA softwareUnited States
Design tools are the entry point to the whole chain: without EDA there is no layout that can be manufactured. That also makes EDA one of the most effective points at which to apply controls.
Advanced lithographyASML, Netherlands
EUV lithography currently has a single supplier: the most extreme concentration in the chain, and the reason the Netherlands has been drawn into US–China disputes over controls.
Materials and certain critical equipmentJapan
Photoresists, speciality chemicals, and several kinds of process equipment. Their share of cost is often tiny, yet without them the line stops; this is where single points of failure most often hide.
Leading-edge process and foundryTaiwan
Separating foundry from design lowered the barrier to entry but concentrated manufacturing risk in a handful of sites. TSMC is expanding 3nm-related capacity in Taiwan, the United States, and Japan, with output ramping up through 2027–2028.
MemoryMainly South Korea
More suppliers than the layers above, yet high-bandwidth memory capacity is just as tight, because AI training throughput depends on memory bandwidth keeping pace.
Packaging, test, and final assemblyChina, Taiwan, and Southeast Asia
The most dispersed and most easily relocated layer, though advanced packaging is moving toward the leading edge and turning part of it into a new bottleneck.

Highly concentrated — few suppliers, not substitutable near-termPartly concentrated — more suppliers, capacity still tightRelatively dispersed — relocatable, but moving toward the leading edge

“The supply chain is concentrated” says little on its own. Concentrated at which layer, in whose hands, and how easily replaced are three separate questions. Laid out layer by layer, the chain shows two things: each of the top four layers has only a few suppliers, and those suppliers sit in different countries; and diversifying design does not offset concentration in manufacturing, because the two are risks at different layers.

The gap a newcomer faces is one of time rather than technology. Subsidies can build fabs but cannot buy yield: a building is a matter of capital, yield a matter of experience.

Source: Impactful Creative, compiled from the division of labour described in our chip-war analysis and the glossary entries on leading-edge process, foundry, and single points of failure

What layers does a chip pass through?

For three decades, semiconductors have been one of globalisation’s most refined divisions of labour: the United States in design and EDA software, Japan in materials and certain critical equipment, the Netherlands’ ASML in advanced lithography, Taiwan in leading-edge processes and foundry manufacturing, South Korea in memory, and China in final assembly and as an enormous market.

The system rests on cost, efficiency, and specialisation, and it has helped drive down the price of smartphones, cloud computing, and consumer electronics.

This is not one production line that can be moved wholesale. It is six layers, each developed separately over decades, and none of them can be replicated just by building a plant.

Why does each layer come down to a handful of firms?

Because at every layer the barriers to entry compound. Equipment sells only once someone is willing to risk a production line to qualify it; a process matures only after enough volume production to bring yields up; materials often take more than a year to certify. The gap a newcomer faces is one of time rather than technology, and the leaders do not stop to wait.

That is why subsidies can build a fab but cannot buy yield. A building is a matter of capital; yield is a matter of experience.

Where do single points of failure usually hide?

Usually not in the most expensive or most visible step. A single point of failure is more often a speciality chemical, a single packaging plant, or a particular inspection tool: a tiny share of the cost, yet without it the line stops.

They are hard to find because they tend to sit with third- or fourth-tier suppliers, while most firms can see only the first tier. Knowing your suppliers does not mean knowing your suppliers’ suppliers.

For the same reason, “we have two suppliers” does not necessarily mean redundancy: both may buy the same material from the same source.

Four things called resilience, and only the first is achievable in the near term

Type of risk	What the problem is	What actually addresses it	What remains afterwards
Geographic concentration“Something happens in one place”	Capacity concentrated in one location, where a single disaster, conflict, or blockade takes all of it out at once.	Building abroad. TSMC is expanding 3nm-related capacity in Taiwan, the United States, and Japan, with output ramping up through 2027–2028.	The layer is just as concentrated. A new site needs the same equipment, materials, and experienced engineers; the location moves, but the bottlenecks upstream and downstream do not. And it takes years to get from announcement to volume production.
Industrial concentration“Only a few firms can do it at all”	A layer with only a handful of suppliers worldwide, because the barriers compound: equipment needs someone willing to qualify it on a real line, a process matures only through volume production, and materials take years to certify.	Nothing works in the short run. Subsidies can build a plant but cannot buy yield: a building is a matter of capital and yield a matter of experience, and the leaders do not stop to wait.	All of it. This is the lasting risk, and activity in the row above routinely masks it, because that row is achievable and this one is not.
Single points of failure buried deepThird- and fourth-tier suppliers	Not in the costly or conspicuous steps but in a speciality chemical, a single packaging plant, or a particular inspection tool: negligible in cost, yet without them the line stops.	Mapping the chain downward rather than spreading outward. The difficulty is that they sit three or four tiers deep, while most firms can see only the first.	Whatever the map missed. Knowing your suppliers does not mean knowing theirs, and no map can confirm its own completeness.
Redundancy that is not“We have two suppliers”	Two suppliers may buy the same material from the same source. The redundancy exists on the vendor list but not in the physical chain.	Tracing back to the common upstream source, rather than counting how many first-tier vendors you have.	This row and the one above are two symptoms of one problem: not seeing, and seeing but miscounting. Neither shows up in a contract or a procurement system.

The distinction that matters most is between the first two rows. Geographic dispersion addresses the risk that something goes wrong in one place; industrial concentration is the risk that only a few firms can do the work at all. The first can be dealt with and the second cannot in the near term; building abroad addresses only the first.

The last two rows are two symptoms of one problem: single points of failure sit three or four tiers deep while most firms see only the first tier, and “we have two suppliers” sounds like redundancy when both may buy the same material from the same source. The right-hand column lists what remains, because remedies listed without their residual risk suggest the problem is solved. In practice, the costliest mistake is to answer the second row with the actions of the first.

Source: Impactful Creative, compiled from the causes of concentration, the usual location of single points of failure, and the timescale of overseas fabs described on this page

Do overseas fabs solve the concentration problem?

They reduce the risk tied to any single site, but they do not make the layer any less concentrated. A new fab still needs the same equipment, the same materials, and the same experienced engineers; the location changes, but the bottlenecks upstream and downstream do not.

Expansion also takes years. TSMC is expanding 3nm-related capacity in Taiwan, the United States, and Japan, with output ramping up through 2027–2028, several years after its second fabs in the United States and Japan were first announced.

Any discussion of resilience has to separate two risks. One is that something goes wrong in a single place, which geographic dispersion can address. The other is industrial concentration: only a few firms in the world can do the work at all, and that cannot be fixed in the near term.

How reliable is the “silicon shield” argument?

The silicon shield argument holds that Taiwan’s irreplaceable position in the chip supply chain is itself a deterrent: destroying that capacity would do grave damage to the world economy, so every side has an incentive to preserve the status quo.

The objection is that the argument assumes rational calculation will outweigh political motive. History has no shortage of costly conflicts, and “very costly” does not mean “will not happen”.

Overseas expansion is read in two ways: as spreading risk, and as diluting the shield. Both readings hold, because they answer different questions. One asks whether the capacity is safe; the other asks whether Taiwan’s bargaining position has changed.

Sources

Semiconductor Industry Association (SIA) and Boston Consulting Group (BCG), Strengthening the Global Semiconductor Supply Chain in an Uncertain Era, Apr 2021
McKinsey & Company, Supply chain risk pulse 2025: Tariffs reshuffle global trade priorities, 2 Dec 2025
Central News Agency (CNA), AI需求強勁驅動台積電成長逾30%　法說會重點一次看, 16 Apr 2026
TSMC, TSMC Announces Updates for TSMC Arizona, 6 Dec 2022
TSMC, JASM Set to Expand in Kumamoto Japan, 6 Feb 2024

Terms used here

Advanced Node — The current leading-edge generation of chip manufacturing, which depends on equipment that only a very small number of suppliers can provide, extreme ultraviolet (EUV) lithography above all. Advanced nodes are the main target of export controls because they combine high technical barriers, extreme geographic concentration, and clear military relevance. Which node the control line falls on often shapes the industry more than whether controls exist at all.
Foundry — The business model in which a manufacturer fabricates chips designed by others and sells no products of its own. Separating fabrication from design lowered entry barriers for chip designers but concentrated the industry’s manufacturing risk in a handful of sites. In resilience terms, diversity in design does not offset concentration in fabrication; the two are risks at different layers.
Single Point of Failure — A node whose failure halts the entire chain. It is not necessarily the most expensive or most visible link; more often it is a specialty chemical, a single packaging plant, or a particular test instrument accounting for a trivial share of cost. Mapping these is hard because they typically sit at tier three or four, while most companies have visibility only to tier one.
Silicon Shield — The proposition that Taiwan’s irreplaceable position in global chip supply is itself a deterrent: disrupting that capacity would damage every major economy, so all parties have an interest in the status quo. The argument is contested because it assumes rational calculation will outweigh political motive, and overseas fab expansion is read both as spreading risk and as diluting the shield.